Re: Exposing PG_VERSION_NUM in pg_config

On Wed, Apr 1, 2015 at 8:50 AM, Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:

> On 3/31/15 1:05 AM, Michael Paquier wrote:
> >
> >
> > On Tue, Mar 31, 2015 at 9:40 AM, Peter Eisentraut <peter_e(at)gmx(dot)net
> > <mailto:peter_e(at)gmx(dot)net>> wrote:
> >
> > On 3/30/15 6:29 PM, Michael Paquier wrote:
> > >
> > >
> > > On Tue, Mar 31, 2015 at 5:39 AM, Peter Eisentraut <peter_e(at)gmx(dot)net
> <mailto:peter_e(at)gmx(dot)net>
> > > <mailto:peter_e(at)gmx(dot)net <mailto:peter_e(at)gmx(dot)net>>> wrote:
> > >
> > > On 3/25/15 1:32 AM, Michael Paquier wrote:
> > > > Well, I have no other cases than ones of the type mentioned
> upthread,
> > > > and honestly I am fine as long as we do not apply maths to a
> version
> > > > string. So attached is a patch that adds VERSION_NUM in
> > > > Makefile.global.
> > >
> > > How would you make use of this in an extension makefile?
> > >
> > >
> > > One use case is regression test list filtering depending on
> backend version.
> >
> > I'm interested in the exact syntax you'd use, to compare it to the
> > currently used techniques.
> >
> >
> > With the presence of VERSION_NUM directly in pg_config, the following
> > expression:
> > VERSION_NUM=$(shell $(PG_CONFIG) --version-num)
>
> My question is, once you have this version number in a variable like
> VERSION_NUM, what specifically would you do with it?
>

For an extension that has a single branch compatible with a set of multiple
major versions of Postgres, the cases are custom values for REGRESS_OPTS
and REGRESS depending on the backend version. I also manipulate on a daily
basis the same set of scripts across many platforms (on Windows as well
using msysgit, and MSVC installation does not have pgxs stuff), so I would
use it to simplify them. It is true that you can already do that by parsing
the output of "pg_config --version", and that I would need to maintain both
versions for some time (custom parsing of pg_config --version, and use of
--version-num if this patch is accepted) but it looks IMO tempting to have
directly the version number thinking long term if there is a simple integer
number available at hand to identify a given version of PG.
--
Michael